search

scikit-learn-contrib / lightning

Large-scale linear classification, regression and ranking in Python
https://contrib.scikit-learn.org/lightning/
1.73k stars 214 forks source link

Custom Penalties and Objectives for all models? #130

Closed arose13 closed 5 years ago

arose13 commented 5 years ago

I cannot tell if this is even a possibility with the current implementation but it would be extremely nice to have such powerful optimizers available for custom convex problems.

vene commented 5 years ago

In general it is quite difficult to have a generic mechanism for plug-and-play penalties, losses and solvers, because the solvers often need different APIs to interact with the problem.

Check out @fabianp's library copt which might offer you more flexibility.

arose13 commented 5 years ago

That's okay, I think the API is so clean as is that further generalizations may make it look ugly.